Comment by MallocVoidstar

That isn't unquantized, it's de-quantized. They went from Q5 to fp16 for use in Pytorch instead of the GGUF ecosystem.